Abstract
When a decision-maker deals with forecasting problems, the multi-variable forecasting methodologies can be adopted to predict the future value of a desired target. However, if the actual values of influential factors related to the desired target are unknown at the time of prediction, the user may have difficulties forecasting the future values of desired target accurately, which may create errors and inaccuracy prediction results. To solve such a problem, this study integrates certain forecasting methods and provides the moving back-propagation neural networks (MBPN), moving fuzzy-neuron networks (MFNN), and Grey-Neurofuzzy method to predict the target value, where actual values of influential factors are unknown at the time of prediction. The demand of critical spare parts in the wafer testing factory was used for the prediction and analysis of forecasting methods. Results indicated that the MBPN, MFNN, and Grey-Neurofuzzy method yield better prediction performance than other forecasting methods mentioned in this study, including the grey prediction, back-propagation neural networks (BPN), and the integration of grey system and neural networks, etc. This study also illustrates the importance of demand forecasting on critical spare parts to maintenance operation. The performance of maintenance indicators were derived and compared; therefore managers can improve the performance of maintenance operation according to the analyzed results.